Study On Content-Based Engineering Drawing Retrieval And Its Key Tchenologies

The reuse of electronic drawings is a effective way to improve the efficiency of CAD software. How to speedily, accurately search out the required drawings from the database is especially important to the user. Nowadays, the query of drawings is performed based on keywords, such as project, designer, drawing number, and so on. The keyword-based query has the merit of quick speed, easy technology realization and use. However, the retrieval way based on keywords has lots of limitations which are consuming more time and labor to label a drawing manually and difficult to completely describe engineering drawings only by keywords. In order to inherit the merit of the keyword-based retrieval way and make up for its shortcomings, this thesis introduces sketching technology to the retrieval of drawing information based on drawing's content. The different between keyword-based retrieval and content-based retrieval is that the later is done by analysis of drawing content to extract the feature of drawing. The query is performed by similarity degree calculation. In the way of content-based retrieval, users'sketchy input can easily find out the expected drawings which structure is basically same as that of users'input. The way of users'sketchy input can greatly decrease the restriction of retrieval condition in existing drawing management system and make the input way more flexibly.Some key techniques of content-based engineering drawing retrieval are studied deeply. The framework of content-based engineering drawing retrieval System (CBEDRS) is proposed and some key techniques, such as, sketch recognition, graphic shape match and relevance feedback are studied. The system prototype is constructed finally.The main contributions of this thesis are the following:(1) The concept of content-based engineering drawing retrieval(CBEDR) is proposed. The thesis introduces the sketch technology to the retrieval of engineering drawing and the concept of content-based engineering drawing retrieval is proposed. The framework of content-based engineering drawing retrieval system is proposed.The system prototype is constructed based on Web Service and N-tiers.(2) A fuzzy recognition method of input sketch is proposedFuzzy set is introduced in CBEDRS to perform sketch recognition. The thesis proposes a fuzzy recognition method according to the maximum membership principle of fuzzy pattern recognition. In the basis of component and simple graphics, the composite graphics is recognized by the relation of topology, orientation and rotation. A method of sketchy input processor , which involves technologies of sketchy input , sketch fuzzy recognition and relevance feedback, is also described in the thesis.(3) A prediction method of users'sketchy input intention is proposed by using Case-based Reasoning(CBR)In order to effectively predict users'sketchy input intention, the thesis stores users'recognition process cases and makes use of distance of fuzzy set to capture users'sketching habits and to reveal user's latent retrieval requirement. Based on the fuzzy representation of sketch recognition process instances, a fuzzy similarity retrieval of sketch recognition instances is described. Finally, a layered fuzzy retrieval method of recognition process instances is also presented, which can improve retrieval efficiency.(4) A similar degree calculation method of graphics based on graph matching algorithm is proposedThe thesis presents a graph-based uniform representation for graphics with different complexity, which converting basis entities respectively into Spatial Relation Graph (SRGs) in different information granularity. Accordingly, the similar degree of graphics can been obtained based on graph matching algorithm. To improve the efficiency of graph matching algorithm, search space is minimized and state space in matching SRGs for drawing retrieval is reduced by predicting the directly next state of the current state and neglect invalid states in order.(5) A relevance feedback method of CBEDR is proposedThe technologies of CBEDR includes explicit feedback in drawing retrieval and implict feedback in sketch recognition. A adaptive adjustment method of threshold value is proposed according to user's relevance judgement for the results. Query point movement and re-weighting algorithm for graphics with single component and multiple components are also presented in the thesis, which improves retrieval accuracy and can effectively capture the user's retrieval intention.